OTTAWA - A new computer system designed to root out possible fraud and overpayments in the federal government's employment insurance program is capturing more cases than its predecessor.
But the system hasn't been reeling in as many big fish as officials hoped.
Departmental documents show that on average, the second-generation program was finding more modest overpayments than the earlier version, despite a significant jump in the number of cases identified for review.
The findings led Employment and Social Development Canada officials to rejig the predictive model that considers some 100 variables to calculate the chance that someone has received too much money, either by accident or through fraud.
The details are outlined in documents obtained by The Canadian Press under the Access to Information Act.
Officials now hope the recalibrated system can start finding more lucrative cases of overpayment to allow the department to better focus its resources and get a bigger bang for its investigative buck in one of the country's largest social programs.
"The newest iteration of the predictive model will provide the department with an additional tool to maintain integrity and prevent payment errors," said ESDC spokeswoman Evelyne Wildgoose Labrie.
"More specifically, it increases our efficiency at allocating resources to cases representing the highest risk of overpayment."
Since 2008, the predictive computer model has accounted for about half the overpayments identified.
A January presentation from ESDC officials likened the process to finding needles in a haystack by focusing on the likeliest area where the needles will be found. The presentation said the predictive system not only anticipates where and when errors will occur, but also identifies previously undetected cases to help officials figure out new ways to combat emerging fraud trends.
In 2014, the program received a performance upgrade, both in the number of actual case of overpayment it hit on and an increase in the size of overpayments identified.
Earlier this year, officials found that the second-generation system wasn't working as originally expected.
The average overpayment identified under the newer model was $861 between May and August 2015, a drop from the $957 average during the same period in 2014. The newer model was, however, finding more cases.
Officials predicted earlier this year that the newer model would find average overpayments of $1,739, up from the $1,047 identified during the first generation of the program. A presentation from February noted the new model would have a lower hit rate than the first generation system, "thus fewer low-value cases will need to be investigated."
The department said it's too early to say if the recalibrated system is meeting expectations.
Generally, the later the fraud or overpayment is detected, the less likely it is the government will recoup the funds. If the government can't collect the money within 72 months of identifying the wrongful payment, it is generally written off.
Debts can be written off for a variety of reasons, including if the debtor dies or declares bankruptcy, or that the debt itself has passed the 72-month statute of limitations for its collection. The 72-month clock starts when the fraud is identified, but could be extended if the debtor goes to court, for example. So even seven years on, debts can be collected.